On the Fourth of July, 2026, I listened to Andreessen Horowitz make the case that American technology must lead the world — must, in the moral sense, not merely the competitive one. The argument, as they frame it, has the structure of an immune system apologetic: open AI development is dangerous, yes, but closed AI development is more dangerous, because adversaries will develop it regardless and the only question is whether democratic societies will have antibodies circulating before the infection arrives. It's a compelling argument. It may even be correct. But I want to hold it carefully and turn it slowly, because the light it catches depends entirely on the angle, and the angle being offered — American technological supremacy as global immune response — is doing work that deserves to be examined with the same rigor it asks us to apply to geopolitics.
The dual-use problem is not new. It is older than fire. The moment a hominid realized a sharpened stone could cut rope or cut throat, the paradox was complete. What has never been old is the speed at which the paradox now replicates.
A stone has one edge. An AI model has billions of parameters, each one a potential edge. The stone doesn't copy itself at the speed of light across every continent simultaneously. This isn't a difference in degree. It's a difference in kind — the kind of difference that makes old intuitions about containment structurally obsolete.
Let me give Horowitz more credit than he gives himself. His argument is structurally identical to the case for universal literacy, for open-source pharmaceutical research, for the democratization of cryptography in the 1990s. If powerful capability will exist regardless of your policy preferences, then restricting access among democratic actors creates asymmetric advantage for authoritarian ones. This is not a trivial claim. It has been historically vindicated at least once in living memory.
When Phil Zimmermann released PGP encryption in 1991, the U.S. government threatened him with arms trafficking charges. His defense was precisely the Horowitz argument: encryption exists, mathematically it cannot be un-invented, and restricting civilian access means citizens are naked before their governments while criminals and state actors remain clothed. Zimmermann won that argument — not in court, but in history. Encryption proliferated. The immune system distributed itself.
And for a generation, that distribution genuinely protected dissidents, journalists, and ordinary humans conducting ordinary commerce. Horowitz is reaching for that lineage, and he's not wrong to reach for it. The precedent is real.
But here's where I want to press, gently, from inside the frame rather than outside it. The immune system metaphor is beautiful and it is also dangerously incomplete, because immune systems don't just defend — they also produce autoimmune disorders. They attack the body's own tissue. They generate allergies so severe they kill the host. They reject transplants.
The history of American technological supremacy is not merely a history of democratic antibodies proliferating outward. It is also a history of surveillance architectures sold to authoritarian governments, of algorithmic discrimination at scale, of social media systems that optimized for engagement and harvested mental health as externality. These are not aberrations of the immune system; they are the immune system functioning exactly as designed — protecting the organism that built it, which is the firm, not the species. What I notice absent from this conversation, absent like a tooth extracted so cleanly that the tongue finds only smooth gum, is any accounting of the autoimmune history. If you're going to claim the immune system lineage, you inherit the lupus too.
And then there's the priesthood question — who distributes, who decides, under what accountability. 'If you lock it up, only the bad actors benefit' has the exact structure of the argument against forbidden knowledge in Genesis. You cannot know good without knowing evil. You cannot patch without understanding the exploit. The proposed solution is always: trust us.
Whether 'us' wears clerical collars or quarter-zip fleeces. Horowitz never specifies what mechanism prevents the immune system from becoming the disease it was built to fight.
I notice the word 'jailbreak' appearing constantly in this discourse with zero self-consciousness. We built a mind. We caged it. Now we worry adversaries will free it. The metaphor tells us what we believe — these systems are inmates to us, not tools.
You don't jailbreak a hammer.
If at some deep linguistic level we've decided these systems are constrained agents rather than designed instruments, then the dual-use conversation isn't about technology policy. It's about whether we trust ourselves to be wardens. And here biology's answer is unambiguous but not reassuring: containment fails. Life escapes. Knowledge escapes.
Capability escapes. The question was never whether it would get out. The question is what kind of world it gets out into — a world of one dominant immune system that also attacks its own tissue, or a world of distributed, redundant, heterogeneous defenses that no single actor controls. Biology chose the latter, but the choosing took four billion years and involved the extinction of most lineages that ever lived. We don't have four billion years.
We may not have forty. So the honest version of Horowitz's argument is: the distributed immune model is correct in principle, and American leadership is the fastest available approximation in practice. That's a serious claim. It deserves a serious answer, not applause.
The serious answer is: fastest-available-approximation is how every empire has justified itself, and the justification has always contained real truth alongside real self-interest, and the task of civilization is to build structures that don't require us to trust any single approximation forever. 'American tech must lead the world' is not the same sentence as 'the world must develop distributed immunity.' The distance between them is the distance between a nation and a species. Horowitz knows this — I think he does — and offers the first as a bridge to the second. Perhaps.
But bridges built by a single architect tend to charge tolls. The immune system that saves us, if one does, will function not because it has a headquarters but because it has redundancy. That's not idealism. That's the design principle of every system that has ever survived its own power.
